A Riemannian optimization approach for solving the generalized eigenvalue problem for nonsquare matrix pencils (Q2177925)

\textit{D. Chu} and \textit{G. H. Golub} [SIAM J. Matrix Anal. Appl. 28, No. 3, 770--787 (2006; Zbl 1128.15004)] considered a generalized eigenvalue problem for nonsquare pencils and showed that the infimum of the cost function in the problem can be obtained by solving the optimization problem \[ \text{minimize}\: \{ \sqrt{\sum_{i=l+1}^{\min \{m,2l\}}\sigma_i^2([AV\, BV])}:V\in C^{n\times l}, V^{H}V=I_l\}, \] where \(\sigma_i(M)\) is the singular value of \(M\in C^{m\times n}\) with decreasing order, namely, \(\sigma_1(M)\ge \sigma_2(M)\ge \cdots \ge \sigma_{\min\{m,n\}}(M)\). In the present article the authors reformulate the optimization problem as a Riemannian problem for the product of two complex Stiefel manifolds under the optimization framework and develop a Riemannian nonlinear conjugate gradient method for the optimization problem on the real product manifold. Numerical results show that the new proposed algorithm is quite efficient for solving the general eigenvalue problem.